2 This file is part of systemd.
4 Copyright 2013 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
23 #include "alloc-util.h"
24 #include "cgroup-util.h"
29 #include "parse-util.h"
30 #include "path-util.h"
31 #include "process-util.h"
32 //#include "special.h"
33 #include "string-table.h"
34 #include "string-util.h"
35 #include "stdio-util.h"
37 #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC)
39 #if 0 /// UNNEEDED by elogind
40 static void cgroup_compat_warn(void) {
41 static bool cgroup_compat_warned = false;
43 if (cgroup_compat_warned)
46 log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. See cgroup-compat debug messages for details.");
47 cgroup_compat_warned = true;
50 #define log_cgroup_compat(unit, fmt, ...) do { \
51 cgroup_compat_warn(); \
52 log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \
55 void cgroup_context_init(CGroupContext *c) {
58 /* Initialize everything to the kernel defaults, assuming the
59 * structure is preinitialized to 0 */
61 c->cpu_weight = CGROUP_WEIGHT_INVALID;
62 c->startup_cpu_weight = CGROUP_WEIGHT_INVALID;
63 c->cpu_quota_per_sec_usec = USEC_INFINITY;
65 c->cpu_shares = CGROUP_CPU_SHARES_INVALID;
66 c->startup_cpu_shares = CGROUP_CPU_SHARES_INVALID;
68 c->memory_high = CGROUP_LIMIT_MAX;
69 c->memory_max = CGROUP_LIMIT_MAX;
70 c->memory_swap_max = CGROUP_LIMIT_MAX;
72 c->memory_limit = CGROUP_LIMIT_MAX;
74 c->io_weight = CGROUP_WEIGHT_INVALID;
75 c->startup_io_weight = CGROUP_WEIGHT_INVALID;
77 c->blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID;
78 c->startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID;
80 c->tasks_max = (uint64_t) -1;
83 void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) {
87 LIST_REMOVE(device_allow, c->device_allow, a);
92 void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) {
96 LIST_REMOVE(device_weights, c->io_device_weights, w);
101 void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) {
105 LIST_REMOVE(device_limits, c->io_device_limits, l);
110 void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) {
114 LIST_REMOVE(device_weights, c->blockio_device_weights, w);
119 void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) {
123 LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b);
128 void cgroup_context_done(CGroupContext *c) {
131 while (c->io_device_weights)
132 cgroup_context_free_io_device_weight(c, c->io_device_weights);
134 while (c->io_device_limits)
135 cgroup_context_free_io_device_limit(c, c->io_device_limits);
137 while (c->blockio_device_weights)
138 cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights);
140 while (c->blockio_device_bandwidths)
141 cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths);
143 while (c->device_allow)
144 cgroup_context_free_device_allow(c, c->device_allow);
146 c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow);
147 c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny);
150 void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) {
151 CGroupIODeviceLimit *il;
152 CGroupIODeviceWeight *iw;
153 CGroupBlockIODeviceBandwidth *b;
154 CGroupBlockIODeviceWeight *w;
155 CGroupDeviceAllow *a;
156 char u[FORMAT_TIMESPAN_MAX];
161 prefix = strempty(prefix);
164 "%sCPUAccounting=%s\n"
165 "%sIOAccounting=%s\n"
166 "%sBlockIOAccounting=%s\n"
167 "%sMemoryAccounting=%s\n"
168 "%sTasksAccounting=%s\n"
169 "%sCPUWeight=%" PRIu64 "\n"
170 "%sStartupCPUWeight=%" PRIu64 "\n"
171 "%sCPUShares=%" PRIu64 "\n"
172 "%sStartupCPUShares=%" PRIu64 "\n"
173 "%sCPUQuotaPerSecSec=%s\n"
174 "%sIOWeight=%" PRIu64 "\n"
175 "%sStartupIOWeight=%" PRIu64 "\n"
176 "%sBlockIOWeight=%" PRIu64 "\n"
177 "%sStartupBlockIOWeight=%" PRIu64 "\n"
178 "%sMemoryLow=%" PRIu64 "\n"
179 "%sMemoryHigh=%" PRIu64 "\n"
180 "%sMemoryMax=%" PRIu64 "\n"
181 "%sMemorySwapMax=%" PRIu64 "\n"
182 "%sMemoryLimit=%" PRIu64 "\n"
183 "%sTasksMax=%" PRIu64 "\n"
184 "%sDevicePolicy=%s\n"
186 prefix, yes_no(c->cpu_accounting),
187 prefix, yes_no(c->io_accounting),
188 prefix, yes_no(c->blockio_accounting),
189 prefix, yes_no(c->memory_accounting),
190 prefix, yes_no(c->tasks_accounting),
191 prefix, c->cpu_weight,
192 prefix, c->startup_cpu_weight,
193 prefix, c->cpu_shares,
194 prefix, c->startup_cpu_shares,
195 prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1),
196 prefix, c->io_weight,
197 prefix, c->startup_io_weight,
198 prefix, c->blockio_weight,
199 prefix, c->startup_blockio_weight,
200 prefix, c->memory_low,
201 prefix, c->memory_high,
202 prefix, c->memory_max,
203 prefix, c->memory_swap_max,
204 prefix, c->memory_limit,
205 prefix, c->tasks_max,
206 prefix, cgroup_device_policy_to_string(c->device_policy),
207 prefix, yes_no(c->delegate));
209 LIST_FOREACH(device_allow, a, c->device_allow)
211 "%sDeviceAllow=%s %s%s%s\n",
214 a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : "");
216 LIST_FOREACH(device_weights, iw, c->io_device_weights)
218 "%sIODeviceWeight=%s %" PRIu64,
223 LIST_FOREACH(device_limits, il, c->io_device_limits) {
224 char buf[FORMAT_BYTES_MAX];
225 CGroupIOLimitType type;
227 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++)
228 if (il->limits[type] != cgroup_io_limit_defaults[type])
232 cgroup_io_limit_type_to_string(type),
234 format_bytes(buf, sizeof(buf), il->limits[type]));
237 LIST_FOREACH(device_weights, w, c->blockio_device_weights)
239 "%sBlockIODeviceWeight=%s %" PRIu64,
244 LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) {
245 char buf[FORMAT_BYTES_MAX];
247 if (b->rbps != CGROUP_LIMIT_MAX)
249 "%sBlockIOReadBandwidth=%s %s\n",
252 format_bytes(buf, sizeof(buf), b->rbps));
253 if (b->wbps != CGROUP_LIMIT_MAX)
255 "%sBlockIOWriteBandwidth=%s %s\n",
258 format_bytes(buf, sizeof(buf), b->wbps));
262 static int lookup_block_device(const char *p, dev_t *dev) {
271 return log_warning_errno(errno, "Couldn't stat device %s: %m", p);
273 if (S_ISBLK(st.st_mode))
275 else if (major(st.st_dev) != 0) {
276 /* If this is not a device node then find the block
277 * device this file is stored on */
280 /* If this is a partition, try to get the originating
282 block_get_whole_disk(*dev, dev);
284 log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p);
291 static int whitelist_device(const char *path, const char *node, const char *acc) {
292 char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4];
294 bool ignore_notfound;
300 if (node[0] == '-') {
301 /* Non-existent paths starting with "-" must be silently ignored */
303 ignore_notfound = true;
305 ignore_notfound = false;
307 if (stat(node, &st) < 0) {
308 if (errno == ENOENT && ignore_notfound)
311 return log_warning_errno(errno, "Couldn't stat device %s: %m", node);
314 if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) {
315 log_warning("%s is not a device.", node);
321 S_ISCHR(st.st_mode) ? 'c' : 'b',
322 major(st.st_rdev), minor(st.st_rdev),
325 r = cg_set_attribute("devices", path, "devices.allow", buf);
327 log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
328 "Failed to set devices.allow on %s: %m", path);
333 static int whitelist_major(const char *path, const char *name, char type, const char *acc) {
334 _cleanup_fclose_ FILE *f = NULL;
341 assert(type == 'b' || type == 'c');
343 f = fopen("/proc/devices", "re");
345 return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type);
347 FOREACH_LINE(line, f, goto fail) {
348 char buf[2+DECIMAL_STR_MAX(unsigned)+3+4], *p, *w;
353 if (type == 'c' && streq(line, "Character devices:")) {
358 if (type == 'b' && streq(line, "Block devices:")) {
373 w = strpbrk(p, WHITESPACE);
378 r = safe_atou(p, &maj);
385 w += strspn(w, WHITESPACE);
387 if (fnmatch(name, w, 0) != 0)
396 r = cg_set_attribute("devices", path, "devices.allow", buf);
398 log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
399 "Failed to set devices.allow on %s: %m", path);
405 return log_warning_errno(errno, "Failed to read /proc/devices: %m");
408 static bool cgroup_context_has_cpu_weight(CGroupContext *c) {
409 return c->cpu_weight != CGROUP_WEIGHT_INVALID ||
410 c->startup_cpu_weight != CGROUP_WEIGHT_INVALID;
413 static bool cgroup_context_has_cpu_shares(CGroupContext *c) {
414 return c->cpu_shares != CGROUP_CPU_SHARES_INVALID ||
415 c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID;
418 static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) {
419 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
420 c->startup_cpu_weight != CGROUP_WEIGHT_INVALID)
421 return c->startup_cpu_weight;
422 else if (c->cpu_weight != CGROUP_WEIGHT_INVALID)
423 return c->cpu_weight;
425 return CGROUP_WEIGHT_DEFAULT;
428 static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) {
429 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
430 c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID)
431 return c->startup_cpu_shares;
432 else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID)
433 return c->cpu_shares;
435 return CGROUP_CPU_SHARES_DEFAULT;
438 static void cgroup_apply_unified_cpu_config(Unit *u, uint64_t weight, uint64_t quota) {
439 char buf[MAX(DECIMAL_STR_MAX(uint64_t) + 1, (DECIMAL_STR_MAX(usec_t) + 1) * 2)];
442 xsprintf(buf, "%" PRIu64 "\n", weight);
443 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.weight", buf);
445 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
446 "Failed to set cpu.weight: %m");
448 if (quota != USEC_INFINITY)
449 xsprintf(buf, USEC_FMT " " USEC_FMT "\n",
450 quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC, CGROUP_CPU_QUOTA_PERIOD_USEC);
452 xsprintf(buf, "max " USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC);
454 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.max", buf);
457 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
458 "Failed to set cpu.max: %m");
461 static void cgroup_apply_legacy_cpu_config(Unit *u, uint64_t shares, uint64_t quota) {
462 char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1];
465 xsprintf(buf, "%" PRIu64 "\n", shares);
466 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.shares", buf);
468 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
469 "Failed to set cpu.shares: %m");
471 xsprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC);
472 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_period_us", buf);
474 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
475 "Failed to set cpu.cfs_period_us: %m");
477 if (quota != USEC_INFINITY) {
478 xsprintf(buf, USEC_FMT "\n", quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC);
479 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", buf);
481 r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", "-1");
483 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
484 "Failed to set cpu.cfs_quota_us: %m");
487 static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) {
488 return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT,
489 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
492 static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) {
493 return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT,
494 CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX);
497 static bool cgroup_context_has_io_config(CGroupContext *c) {
498 return c->io_accounting ||
499 c->io_weight != CGROUP_WEIGHT_INVALID ||
500 c->startup_io_weight != CGROUP_WEIGHT_INVALID ||
501 c->io_device_weights ||
505 static bool cgroup_context_has_blockio_config(CGroupContext *c) {
506 return c->blockio_accounting ||
507 c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID ||
508 c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID ||
509 c->blockio_device_weights ||
510 c->blockio_device_bandwidths;
513 static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) {
514 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
515 c->startup_io_weight != CGROUP_WEIGHT_INVALID)
516 return c->startup_io_weight;
517 else if (c->io_weight != CGROUP_WEIGHT_INVALID)
520 return CGROUP_WEIGHT_DEFAULT;
523 static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) {
524 if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) &&
525 c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID)
526 return c->startup_blockio_weight;
527 else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID)
528 return c->blockio_weight;
530 return CGROUP_BLKIO_WEIGHT_DEFAULT;
533 static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) {
534 return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT,
535 CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX);
538 static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) {
539 return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT,
540 CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX);
543 static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) {
544 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
548 r = lookup_block_device(dev_path, &dev);
552 xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight);
553 r = cg_set_attribute("io", u->cgroup_path, "io.weight", buf);
555 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
556 "Failed to set io.weight: %m");
559 static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) {
560 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
564 r = lookup_block_device(dev_path, &dev);
568 xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight);
569 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.weight_device", buf);
571 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
572 "Failed to set blkio.weight_device: %m");
575 static unsigned cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) {
576 char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)];
577 char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4];
578 CGroupIOLimitType type;
583 r = lookup_block_device(dev_path, &dev);
587 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) {
588 if (limits[type] != cgroup_io_limit_defaults[type]) {
589 xsprintf(limit_bufs[type], "%" PRIu64, limits[type]);
592 xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0");
596 xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev),
597 limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX],
598 limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]);
599 r = cg_set_attribute("io", u->cgroup_path, "io.max", buf);
601 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
602 "Failed to set io.max: %m");
606 static unsigned cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) {
607 char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1];
612 r = lookup_block_device(dev_path, &dev);
616 if (rbps != CGROUP_LIMIT_MAX)
618 sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps);
619 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.read_bps_device", buf);
621 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
622 "Failed to set blkio.throttle.read_bps_device: %m");
624 if (wbps != CGROUP_LIMIT_MAX)
626 sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps);
627 r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.write_bps_device", buf);
629 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
630 "Failed to set blkio.throttle.write_bps_device: %m");
635 static bool cgroup_context_has_unified_memory_config(CGroupContext *c) {
636 return c->memory_low > 0 || c->memory_high != CGROUP_LIMIT_MAX || c->memory_max != CGROUP_LIMIT_MAX || c->memory_swap_max != CGROUP_LIMIT_MAX;
639 static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) {
640 char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max";
643 if (v != CGROUP_LIMIT_MAX)
644 xsprintf(buf, "%" PRIu64 "\n", v);
646 r = cg_set_attribute("memory", u->cgroup_path, file, buf);
648 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
649 "Failed to set %s: %m", file);
652 static void cgroup_context_apply(Unit *u, CGroupMask mask, ManagerState state) {
660 c = unit_get_cgroup_context(u);
661 path = u->cgroup_path;
669 /* Some cgroup attributes are not supported on the root cgroup,
670 * hence silently ignore */
671 is_root = isempty(path) || path_equal(path, "/");
673 /* Make sure we don't try to display messages with an empty path. */
676 /* We generally ignore errors caused by read-only mounted
677 * cgroup trees (assuming we are running in a container then),
678 * and missing cgroups, i.e. EROFS and ENOENT. */
680 if ((mask & CGROUP_MASK_CPU) && !is_root) {
681 bool has_weight = cgroup_context_has_cpu_weight(c);
682 bool has_shares = cgroup_context_has_cpu_shares(c);
684 if (cg_all_unified() > 0) {
688 weight = cgroup_context_cpu_weight(c, state);
689 else if (has_shares) {
690 uint64_t shares = cgroup_context_cpu_shares(c, state);
692 weight = cgroup_cpu_shares_to_weight(shares);
694 log_cgroup_compat(u, "Applying [Startup]CpuShares %" PRIu64 " as [Startup]CpuWeight %" PRIu64 " on %s",
695 shares, weight, path);
697 weight = CGROUP_WEIGHT_DEFAULT;
699 cgroup_apply_unified_cpu_config(u, weight, c->cpu_quota_per_sec_usec);
704 uint64_t weight = cgroup_context_cpu_weight(c, state);
706 shares = cgroup_cpu_weight_to_shares(weight);
708 log_cgroup_compat(u, "Applying [Startup]CpuWeight %" PRIu64 " as [Startup]CpuShares %" PRIu64 " on %s",
709 weight, shares, path);
710 } else if (has_shares)
711 shares = cgroup_context_cpu_shares(c, state);
713 shares = CGROUP_CPU_SHARES_DEFAULT;
715 cgroup_apply_legacy_cpu_config(u, shares, c->cpu_quota_per_sec_usec);
719 if (mask & CGROUP_MASK_IO) {
720 bool has_io = cgroup_context_has_io_config(c);
721 bool has_blockio = cgroup_context_has_blockio_config(c);
724 char buf[8+DECIMAL_STR_MAX(uint64_t)+1];
728 weight = cgroup_context_io_weight(c, state);
729 else if (has_blockio) {
730 uint64_t blkio_weight = cgroup_context_blkio_weight(c, state);
732 weight = cgroup_weight_blkio_to_io(blkio_weight);
734 log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64,
735 blkio_weight, weight);
737 weight = CGROUP_WEIGHT_DEFAULT;
739 xsprintf(buf, "default %" PRIu64 "\n", weight);
740 r = cg_set_attribute("io", path, "io.weight", buf);
742 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
743 "Failed to set io.weight: %m");
746 CGroupIODeviceWeight *w;
748 /* FIXME: no way to reset this list */
749 LIST_FOREACH(device_weights, w, c->io_device_weights)
750 cgroup_apply_io_device_weight(u, w->path, w->weight);
751 } else if (has_blockio) {
752 CGroupBlockIODeviceWeight *w;
754 /* FIXME: no way to reset this list */
755 LIST_FOREACH(device_weights, w, c->blockio_device_weights) {
756 weight = cgroup_weight_blkio_to_io(w->weight);
758 log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s",
759 w->weight, weight, w->path);
761 cgroup_apply_io_device_weight(u, w->path, weight);
766 /* Apply limits and free ones without config. */
768 CGroupIODeviceLimit *l, *next;
770 LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) {
771 if (!cgroup_apply_io_device_limit(u, l->path, l->limits))
772 cgroup_context_free_io_device_limit(c, l);
774 } else if (has_blockio) {
775 CGroupBlockIODeviceBandwidth *b, *next;
777 LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) {
778 uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX];
779 CGroupIOLimitType type;
781 for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++)
782 limits[type] = cgroup_io_limit_defaults[type];
784 limits[CGROUP_IO_RBPS_MAX] = b->rbps;
785 limits[CGROUP_IO_WBPS_MAX] = b->wbps;
787 log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s",
788 b->rbps, b->wbps, b->path);
790 if (!cgroup_apply_io_device_limit(u, b->path, limits))
791 cgroup_context_free_blockio_device_bandwidth(c, b);
796 if (mask & CGROUP_MASK_BLKIO) {
797 bool has_io = cgroup_context_has_io_config(c);
798 bool has_blockio = cgroup_context_has_blockio_config(c);
801 char buf[DECIMAL_STR_MAX(uint64_t)+1];
805 uint64_t io_weight = cgroup_context_io_weight(c, state);
807 weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state));
809 log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64,
811 } else if (has_blockio)
812 weight = cgroup_context_blkio_weight(c, state);
814 weight = CGROUP_BLKIO_WEIGHT_DEFAULT;
816 xsprintf(buf, "%" PRIu64 "\n", weight);
817 r = cg_set_attribute("blkio", path, "blkio.weight", buf);
819 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
820 "Failed to set blkio.weight: %m");
823 CGroupIODeviceWeight *w;
825 /* FIXME: no way to reset this list */
826 LIST_FOREACH(device_weights, w, c->io_device_weights) {
827 weight = cgroup_weight_io_to_blkio(w->weight);
829 log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s",
830 w->weight, weight, w->path);
832 cgroup_apply_blkio_device_weight(u, w->path, weight);
834 } else if (has_blockio) {
835 CGroupBlockIODeviceWeight *w;
837 /* FIXME: no way to reset this list */
838 LIST_FOREACH(device_weights, w, c->blockio_device_weights)
839 cgroup_apply_blkio_device_weight(u, w->path, w->weight);
843 /* Apply limits and free ones without config. */
845 CGroupIODeviceLimit *l, *next;
847 LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) {
848 log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s",
849 l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path);
851 if (!cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]))
852 cgroup_context_free_io_device_limit(c, l);
854 } else if (has_blockio) {
855 CGroupBlockIODeviceBandwidth *b, *next;
857 LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths)
858 if (!cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps))
859 cgroup_context_free_blockio_device_bandwidth(c, b);
863 if ((mask & CGROUP_MASK_MEMORY) && !is_root) {
864 if (cg_all_unified() > 0) {
865 uint64_t max, swap_max = CGROUP_LIMIT_MAX;
867 if (cgroup_context_has_unified_memory_config(c)) {
869 swap_max = c->memory_swap_max;
871 max = c->memory_limit;
873 if (max != CGROUP_LIMIT_MAX)
874 log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max);
877 cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low);
878 cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high);
879 cgroup_apply_unified_memory_limit(u, "memory.max", max);
880 cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max);
882 char buf[DECIMAL_STR_MAX(uint64_t) + 1];
885 if (cgroup_context_has_unified_memory_config(c)) {
887 log_cgroup_compat(u, "Applying MemoryMax %" PRIi64 " as MemoryLimit", val);
889 val = c->memory_limit;
891 if (val == CGROUP_LIMIT_MAX)
892 strncpy(buf, "-1\n", sizeof(buf));
894 xsprintf(buf, "%" PRIu64 "\n", val);
896 r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf);
898 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
899 "Failed to set memory.limit_in_bytes: %m");
903 if ((mask & CGROUP_MASK_DEVICES) && !is_root) {
904 CGroupDeviceAllow *a;
906 /* Changing the devices list of a populated cgroup
907 * might result in EINVAL, hence ignore EINVAL
910 if (c->device_allow || c->device_policy != CGROUP_AUTO)
911 r = cg_set_attribute("devices", path, "devices.deny", "a");
913 r = cg_set_attribute("devices", path, "devices.allow", "a");
915 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
916 "Failed to reset devices.list: %m");
918 if (c->device_policy == CGROUP_CLOSED ||
919 (c->device_policy == CGROUP_AUTO && c->device_allow)) {
920 static const char auto_devices[] =
921 "/dev/null\0" "rwm\0"
922 "/dev/zero\0" "rwm\0"
923 "/dev/full\0" "rwm\0"
924 "/dev/random\0" "rwm\0"
925 "/dev/urandom\0" "rwm\0"
927 "/dev/pts/ptmx\0" "rw\0" /* /dev/pts/ptmx may not be duplicated, but accessed */
928 /* Allow /run/elogind/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */
929 /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */
930 "-/run/systemd/inaccessible/chr\0" "rwm\0"
931 "-/run/systemd/inaccessible/blk\0" "rwm\0";
935 NULSTR_FOREACH_PAIR(x, y, auto_devices)
936 whitelist_device(path, x, y);
938 whitelist_major(path, "pts", 'c', "rw");
941 LIST_FOREACH(device_allow, a, c->device_allow) {
957 if (path_startswith(a->path, "/dev/"))
958 whitelist_device(path, a->path, acc);
959 else if ((val = startswith(a->path, "block-")))
960 whitelist_major(path, val, 'b', acc);
961 else if ((val = startswith(a->path, "char-")))
962 whitelist_major(path, val, 'c', acc);
964 log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path);
968 if ((mask & CGROUP_MASK_PIDS) && !is_root) {
970 if (c->tasks_max != CGROUP_LIMIT_MAX) {
971 char buf[DECIMAL_STR_MAX(uint64_t) + 2];
973 sprintf(buf, "%" PRIu64 "\n", c->tasks_max);
974 r = cg_set_attribute("pids", path, "pids.max", buf);
976 r = cg_set_attribute("pids", path, "pids.max", "max");
979 log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r,
980 "Failed to set pids.max: %m");
984 CGroupMask cgroup_context_get_mask(CGroupContext *c) {
987 /* Figure out which controllers we need */
989 if (c->cpu_accounting ||
990 cgroup_context_has_cpu_weight(c) ||
991 cgroup_context_has_cpu_shares(c) ||
992 c->cpu_quota_per_sec_usec != USEC_INFINITY)
993 mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU;
995 if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c))
996 mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO;
998 if (c->memory_accounting ||
999 c->memory_limit != CGROUP_LIMIT_MAX ||
1000 cgroup_context_has_unified_memory_config(c))
1001 mask |= CGROUP_MASK_MEMORY;
1003 if (c->device_allow ||
1004 c->device_policy != CGROUP_AUTO)
1005 mask |= CGROUP_MASK_DEVICES;
1007 if (c->tasks_accounting ||
1008 c->tasks_max != (uint64_t) -1)
1009 mask |= CGROUP_MASK_PIDS;
1014 CGroupMask unit_get_own_mask(Unit *u) {
1017 /* Returns the mask of controllers the unit needs for itself */
1019 c = unit_get_cgroup_context(u);
1023 /* If delegation is turned on, then turn on all cgroups,
1024 * unless we are on the legacy hierarchy and the process we
1025 * fork into it is known to drop privileges, and hence
1026 * shouldn't get access to the controllers.
1028 * Note that on the unified hierarchy it is safe to delegate
1029 * controllers to unprivileged services. */
1034 e = unit_get_exec_context(u);
1036 exec_context_maintains_privileges(e) ||
1037 cg_all_unified() > 0)
1038 return _CGROUP_MASK_ALL;
1041 return cgroup_context_get_mask(c);
1044 CGroupMask unit_get_members_mask(Unit *u) {
1047 /* Returns the mask of controllers all of the unit's children
1048 * require, merged */
1050 if (u->cgroup_members_mask_valid)
1051 return u->cgroup_members_mask;
1053 u->cgroup_members_mask = 0;
1055 if (u->type == UNIT_SLICE) {
1059 SET_FOREACH(member, u->dependencies[UNIT_BEFORE], i) {
1064 if (UNIT_DEREF(member->slice) != u)
1067 u->cgroup_members_mask |=
1068 unit_get_own_mask(member) |
1069 unit_get_members_mask(member);
1073 u->cgroup_members_mask_valid = true;
1074 return u->cgroup_members_mask;
1077 CGroupMask unit_get_siblings_mask(Unit *u) {
1080 /* Returns the mask of controllers all of the unit's siblings
1081 * require, i.e. the members mask of the unit's parent slice
1082 * if there is one. */
1084 if (UNIT_ISSET(u->slice))
1085 return unit_get_members_mask(UNIT_DEREF(u->slice));
1087 return unit_get_own_mask(u) | unit_get_members_mask(u);
1090 CGroupMask unit_get_subtree_mask(Unit *u) {
1092 /* Returns the mask of this subtree, meaning of the group
1093 * itself and its children. */
1095 return unit_get_own_mask(u) | unit_get_members_mask(u);
1098 CGroupMask unit_get_target_mask(Unit *u) {
1101 /* This returns the cgroup mask of all controllers to enable
1102 * for a specific cgroup, i.e. everything it needs itself,
1103 * plus all that its children need, plus all that its siblings
1104 * need. This is primarily useful on the legacy cgroup
1105 * hierarchy, where we need to duplicate each cgroup in each
1106 * hierarchy that shall be enabled for it. */
1108 mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u);
1109 mask &= u->manager->cgroup_supported;
1114 CGroupMask unit_get_enable_mask(Unit *u) {
1117 /* This returns the cgroup mask of all controllers to enable
1118 * for the children of a specific cgroup. This is primarily
1119 * useful for the unified cgroup hierarchy, where each cgroup
1120 * controls which controllers are enabled for its children. */
1122 mask = unit_get_members_mask(u);
1123 mask &= u->manager->cgroup_supported;
1128 /* Recurse from a unit up through its containing slices, propagating
1129 * mask bits upward. A unit is also member of itself. */
1130 void unit_update_cgroup_members_masks(Unit *u) {
1136 /* Calculate subtree mask */
1137 m = unit_get_subtree_mask(u);
1139 /* See if anything changed from the previous invocation. If
1140 * not, we're done. */
1141 if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask)
1145 u->cgroup_subtree_mask_valid &&
1146 ((m & ~u->cgroup_subtree_mask) != 0) &&
1147 ((~m & u->cgroup_subtree_mask) == 0);
1149 u->cgroup_subtree_mask = m;
1150 u->cgroup_subtree_mask_valid = true;
1152 if (UNIT_ISSET(u->slice)) {
1153 Unit *s = UNIT_DEREF(u->slice);
1156 /* There's more set now than before. We
1157 * propagate the new mask to the parent's mask
1158 * (not caring if it actually was valid or
1161 s->cgroup_members_mask |= m;
1164 /* There's less set now than before (or we
1165 * don't know), we need to recalculate
1166 * everything, so let's invalidate the
1167 * parent's members mask */
1169 s->cgroup_members_mask_valid = false;
1171 /* And now make sure that this change also hits our
1173 unit_update_cgroup_members_masks(s);
1177 static const char *migrate_callback(CGroupMask mask, void *userdata) {
1184 if (u->cgroup_path &&
1185 u->cgroup_realized &&
1186 (u->cgroup_realized_mask & mask) == mask)
1187 return u->cgroup_path;
1189 u = UNIT_DEREF(u->slice);
1195 char *unit_default_cgroup_path(Unit *u) {
1196 _cleanup_free_ char *escaped = NULL, *slice = NULL;
1201 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1202 return strdup(u->manager->cgroup_root);
1204 if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) {
1205 r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice);
1210 escaped = cg_escape(u->id);
1215 return strjoin(u->manager->cgroup_root, "/", slice, "/",
1218 return strjoin(u->manager->cgroup_root, "/", escaped);
1221 int unit_set_cgroup_path(Unit *u, const char *path) {
1222 _cleanup_free_ char *p = NULL;
1234 if (streq_ptr(u->cgroup_path, p))
1238 r = hashmap_put(u->manager->cgroup_unit, p, u);
1243 unit_release_cgroup(u);
1251 int unit_watch_cgroup(Unit *u) {
1252 _cleanup_free_ char *events = NULL;
1257 if (!u->cgroup_path)
1260 if (u->cgroup_inotify_wd >= 0)
1263 /* Only applies to the unified hierarchy */
1264 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1266 return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m");
1270 /* Don't watch the root slice, it's pointless. */
1271 if (unit_has_name(u, SPECIAL_ROOT_SLICE))
1274 r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops);
1278 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events);
1282 u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY);
1283 if (u->cgroup_inotify_wd < 0) {
1285 if (errno == ENOENT) /* If the directory is already
1286 * gone we don't need to track
1287 * it, so this is not an error */
1290 return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path);
1293 r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u);
1295 return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m");
1300 static int unit_create_cgroup(
1302 CGroupMask target_mask,
1303 CGroupMask enable_mask) {
1310 c = unit_get_cgroup_context(u);
1314 if (!u->cgroup_path) {
1315 _cleanup_free_ char *path = NULL;
1317 path = unit_default_cgroup_path(u);
1321 r = unit_set_cgroup_path(u, path);
1323 return log_unit_error_errno(u, r, "Control group %s exists already.", path);
1325 return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path);
1328 /* First, create our own group */
1329 r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path);
1331 return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path);
1333 /* Start watching it */
1334 (void) unit_watch_cgroup(u);
1336 /* Enable all controllers we need */
1337 r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path);
1339 log_unit_warning_errno(u, r, "Failed to enable controllers on cgroup %s, ignoring: %m", u->cgroup_path);
1341 /* Keep track that this is now realized */
1342 u->cgroup_realized = true;
1343 u->cgroup_realized_mask = target_mask;
1344 u->cgroup_enabled_mask = enable_mask;
1346 if (u->type != UNIT_SLICE && !c->delegate) {
1348 /* Then, possibly move things over, but not if
1349 * subgroups may contain processes, which is the case
1350 * for slice and delegation units. */
1351 r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u);
1353 log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path);
1359 int unit_attach_pids_to_cgroup(Unit *u) {
1363 r = unit_realize_cgroup(u);
1367 r = cg_attach_many_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->pids, migrate_callback, u);
1374 static void cgroup_xattr_apply(Unit *u) {
1375 char ids[SD_ID128_STRING_MAX];
1380 if (!MANAGER_IS_SYSTEM(u->manager))
1383 if (sd_id128_is_null(u->invocation_id))
1386 r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path,
1387 "trusted.invocation_id",
1388 sd_id128_to_string(u->invocation_id, ids), 32,
1391 log_unit_warning_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path);
1394 static bool unit_has_mask_realized(Unit *u, CGroupMask target_mask, CGroupMask enable_mask) {
1397 return u->cgroup_realized && u->cgroup_realized_mask == target_mask && u->cgroup_enabled_mask == enable_mask;
1400 /* Check if necessary controllers and attributes for a unit are in place.
1402 * If so, do nothing.
1403 * If not, create paths, move processes over, and set attributes.
1405 * Returns 0 on success and < 0 on failure. */
1406 static int unit_realize_cgroup_now(Unit *u, ManagerState state) {
1407 CGroupMask target_mask, enable_mask;
1412 if (u->in_cgroup_queue) {
1413 LIST_REMOVE(cgroup_queue, u->manager->cgroup_queue, u);
1414 u->in_cgroup_queue = false;
1417 target_mask = unit_get_target_mask(u);
1418 enable_mask = unit_get_enable_mask(u);
1420 if (unit_has_mask_realized(u, target_mask, enable_mask))
1423 /* First, realize parents */
1424 if (UNIT_ISSET(u->slice)) {
1425 r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state);
1430 /* And then do the real work */
1431 r = unit_create_cgroup(u, target_mask, enable_mask);
1435 /* Finally, apply the necessary attributes. */
1436 cgroup_context_apply(u, target_mask, state);
1437 cgroup_xattr_apply(u);
1442 static void unit_add_to_cgroup_queue(Unit *u) {
1444 if (u->in_cgroup_queue)
1447 LIST_PREPEND(cgroup_queue, u->manager->cgroup_queue, u);
1448 u->in_cgroup_queue = true;
1451 unsigned manager_dispatch_cgroup_queue(Manager *m) {
1457 state = manager_state(m);
1459 while ((i = m->cgroup_queue)) {
1460 assert(i->in_cgroup_queue);
1462 r = unit_realize_cgroup_now(i, state);
1464 log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id);
1472 static void unit_queue_siblings(Unit *u) {
1475 /* This adds the siblings of the specified unit and the
1476 * siblings of all parent units to the cgroup queue. (But
1477 * neither the specified unit itself nor the parents.) */
1479 while ((slice = UNIT_DEREF(u->slice))) {
1483 SET_FOREACH(m, slice->dependencies[UNIT_BEFORE], i) {
1487 /* Skip units that have a dependency on the slice
1488 * but aren't actually in it. */
1489 if (UNIT_DEREF(m->slice) != slice)
1492 /* No point in doing cgroup application for units
1493 * without active processes. */
1494 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m)))
1497 /* If the unit doesn't need any new controllers
1498 * and has current ones realized, it doesn't need
1500 if (unit_has_mask_realized(m, unit_get_target_mask(m), unit_get_enable_mask(m)))
1503 unit_add_to_cgroup_queue(m);
1510 int unit_realize_cgroup(Unit *u) {
1513 if (!UNIT_HAS_CGROUP_CONTEXT(u))
1516 /* So, here's the deal: when realizing the cgroups for this
1517 * unit, we need to first create all parents, but there's more
1518 * actually: for the weight-based controllers we also need to
1519 * make sure that all our siblings (i.e. units that are in the
1520 * same slice as we are) have cgroups, too. Otherwise, things
1521 * would become very uneven as each of their processes would
1522 * get as much resources as all our group together. This call
1523 * will synchronously create the parent cgroups, but will
1524 * defer work on the siblings to the next event loop
1527 /* Add all sibling slices to the cgroup queue. */
1528 unit_queue_siblings(u);
1530 /* And realize this one now (and apply the values) */
1531 return unit_realize_cgroup_now(u, manager_state(u->manager));
1534 void unit_release_cgroup(Unit *u) {
1537 /* Forgets all cgroup details for this cgroup */
1539 if (u->cgroup_path) {
1540 (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path);
1541 u->cgroup_path = mfree(u->cgroup_path);
1544 if (u->cgroup_inotify_wd >= 0) {
1545 if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0)
1546 log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring", u->cgroup_inotify_wd, u->id);
1548 (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd));
1549 u->cgroup_inotify_wd = -1;
1553 void unit_prune_cgroup(Unit *u) {
1559 /* Removes the cgroup, if empty and possible, and stops watching it. */
1561 if (!u->cgroup_path)
1564 (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */
1566 is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE);
1568 r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice);
1570 log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path);
1577 unit_release_cgroup(u);
1579 u->cgroup_realized = false;
1580 u->cgroup_realized_mask = 0;
1581 u->cgroup_enabled_mask = 0;
1584 int unit_search_main_pid(Unit *u, pid_t *ret) {
1585 _cleanup_fclose_ FILE *f = NULL;
1586 pid_t pid = 0, npid, mypid;
1592 if (!u->cgroup_path)
1595 r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f);
1599 mypid = getpid_cached();
1600 while (cg_read_pid(f, &npid) > 0) {
1606 /* Ignore processes that aren't our kids */
1607 if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid)
1611 /* Dang, there's more than one daemonized PID
1612 in this group, so we don't know what process
1613 is the main process. */
1624 static int unit_watch_pids_in_path(Unit *u, const char *path) {
1625 _cleanup_closedir_ DIR *d = NULL;
1626 _cleanup_fclose_ FILE *f = NULL;
1632 r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f);
1638 while ((r = cg_read_pid(f, &pid)) > 0) {
1639 r = unit_watch_pid(u, pid);
1640 if (r < 0 && ret >= 0)
1644 if (r < 0 && ret >= 0)
1648 r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d);
1655 while ((r = cg_read_subgroup(d, &fn)) > 0) {
1656 _cleanup_free_ char *p = NULL;
1658 p = strjoin(path, "/", fn);
1664 r = unit_watch_pids_in_path(u, p);
1665 if (r < 0 && ret >= 0)
1669 if (r < 0 && ret >= 0)
1676 int unit_watch_all_pids(Unit *u) {
1681 /* Adds all PIDs from our cgroup to the set of PIDs we
1682 * watch. This is a fallback logic for cases where we do not
1683 * get reliable cgroup empty notifications: we try to use
1684 * SIGCHLD as replacement. */
1686 if (!u->cgroup_path)
1689 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1692 if (r > 0) /* On unified we can use proper notifications */
1695 return unit_watch_pids_in_path(u, u->cgroup_path);
1698 int unit_notify_cgroup_empty(Unit *u) {
1703 if (!u->cgroup_path)
1706 r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path);
1710 unit_add_to_gc_queue(u);
1712 if (UNIT_VTABLE(u)->notify_cgroup_empty)
1713 UNIT_VTABLE(u)->notify_cgroup_empty(u);
1718 static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
1719 Manager *m = userdata;
1726 union inotify_event_buffer buffer;
1727 struct inotify_event *e;
1730 l = read(fd, &buffer, sizeof(buffer));
1732 if (errno == EINTR || errno == EAGAIN)
1735 return log_error_errno(errno, "Failed to read control group inotify events: %m");
1738 FOREACH_INOTIFY_EVENT(e, buffer, l) {
1742 /* Queue overflow has no watch descriptor */
1745 if (e->mask & IN_IGNORED)
1746 /* The watch was just removed */
1749 u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd));
1750 if (!u) /* Not that inotify might deliver
1751 * events for a watch even after it
1752 * was removed, because it was queued
1753 * before the removal. Let's ignore
1754 * this here safely. */
1757 (void) unit_notify_cgroup_empty(u);
1763 int manager_setup_cgroup(Manager *m) {
1764 _cleanup_free_ char *path = NULL;
1765 const char *scope_path;
1772 /* 1. Determine hierarchy */
1773 m->cgroup_root = mfree(m->cgroup_root);
1774 #if 0 /// elogind is not init and must therefore search for PID 1 instead of self.
1775 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root);
1777 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 1, &m->cgroup_root);
1780 return log_error_errno(r, "Cannot determine cgroup we are running in: %m");
1782 #if 0 /// elogind does not support systemd scopes and slices
1783 /* Chop off the init scope, if we are already located in it */
1784 e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE);
1786 /* LEGACY: Also chop off the system slice if we are in
1787 * it. This is to support live upgrades from older systemd
1788 * versions where PID 1 was moved there. Also see
1789 * cg_get_root_path(). */
1790 if (!e && MANAGER_IS_SYSTEM(m)) {
1791 e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE);
1793 e = endswith(m->cgroup_root, "/system"); /* even more legacy */
1799 /* And make sure to store away the root value without trailing
1800 * slash, even for the root dir, so that we can easily prepend
1802 while ((e = endswith(m->cgroup_root, "/")))
1804 log_debug_elogind("Cgroup Controller \"%s\" -> root \"%s\"",
1805 SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root);
1808 r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path);
1810 return log_error_errno(r, "Cannot find cgroup mount point: %m");
1812 r = cg_unified_flush();
1814 return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m");
1816 all_unified = cg_all_unified();
1818 return log_error_errno(r, "Couldn't determine whether we are in all unified mode: %m");
1820 log_debug("Unified cgroup hierarchy is located at %s.", path);
1822 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1824 return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m");
1826 log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path);
1828 log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path);
1831 #if 0 /// elogind is not init, and does not install the agent here.
1832 /* 3. Install agent */
1833 if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) {
1835 /* In the unified hierarchy we can get
1836 * cgroup empty notifications via inotify. */
1838 m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source);
1839 safe_close(m->cgroup_inotify_fd);
1841 m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
1842 if (m->cgroup_inotify_fd < 0)
1843 return log_error_errno(errno, "Failed to create control group inotify object: %m");
1845 r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m);
1847 return log_error_errno(r, "Failed to watch control group inotify object: %m");
1849 /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also
1850 * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */
1851 r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-5);
1853 return log_error_errno(r, "Failed to set priority of inotify event source: %m");
1855 (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify");
1857 } else if (MANAGER_IS_SYSTEM(m) && m->test_run_flags == 0) {
1859 /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable,
1860 * since it does not generate events when control groups with children run empty. */
1862 r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH);
1864 log_warning_errno(r, "Failed to install release agent, ignoring: %m");
1866 log_debug("Installed release agent.");
1868 log_debug("Release agent already installed.");
1871 /* 4. Make sure we are in the special "init.scope" unit in the root slice. */
1872 scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE);
1873 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
1876 * This method is in core, and normally called by systemd
1877 * being init. As elogind is never init, we can not install
1878 * our agent here. We do so when mounting our cgroup file
1879 * system, so only if elogind is its own tiny controller.
1880 * Further, elogind is not meant to run in systemd init scope. */
1881 if (MANAGER_IS_SYSTEM(m))
1882 // we are our own cgroup controller
1883 scope_path = strjoina("");
1884 else if (streq(m->cgroup_root, "/elogind"))
1885 // root already is our cgroup
1886 scope_path = strjoina(m->cgroup_root);
1888 // we have to create our own group
1889 scope_path = strjoina(m->cgroup_root, "/elogind");
1890 r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
1893 return log_error_errno(r, "Failed to create %s control group: %m", scope_path);
1894 log_debug_elogind("Created control group \"%s\"", scope_path);
1896 #if 0 /// elogind is not a "sub-controller" like systemd, so migration is not needed.
1897 /* also, move all other userspace processes remaining
1898 * in the root cgroup into that scope. */
1899 r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0);
1901 log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m");
1904 /* 5. And pin it, so that it cannot be unmounted */
1905 safe_close(m->pin_cgroupfs_fd);
1906 m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK);
1907 if (m->pin_cgroupfs_fd < 0)
1908 return log_error_errno(errno, "Failed to open pin file: %m");
1910 /* 6. Always enable hierarchical support if it exists... */
1911 if (!all_unified && m->test_run_flags == 0)
1912 (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1");
1914 /* 7. Figure out which controllers are supported */
1915 r = cg_mask_supported(&m->cgroup_supported);
1917 return log_error_errno(r, "Failed to determine supported controllers: %m");
1919 for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++)
1920 log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c)));
1925 void manager_shutdown_cgroup(Manager *m, bool delete) {
1928 /* We can't really delete the group, since we are in it. But
1930 if (delete && m->cgroup_root)
1931 (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false);
1933 #if 0 /// elogind does not support the unified hierarchy, yet.
1934 m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit);
1936 m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source);
1937 m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd);
1940 m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd);
1942 m->cgroup_root = mfree(m->cgroup_root);
1945 #if 0 /// UNNEEDED by elogind
1946 Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) {
1953 u = hashmap_get(m->cgroup_unit, cgroup);
1957 p = strdupa(cgroup);
1961 e = strrchr(p, '/');
1963 return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE);
1967 u = hashmap_get(m->cgroup_unit, p);
1973 Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) {
1974 _cleanup_free_ char *cgroup = NULL;
1982 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup);
1986 return manager_get_unit_by_cgroup(m, cgroup);
1989 Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) {
1998 return hashmap_get(m->units, SPECIAL_INIT_SCOPE);
2000 u = hashmap_get(m->watch_pids1, PID_TO_PTR(pid));
2004 u = hashmap_get(m->watch_pids2, PID_TO_PTR(pid));
2008 return manager_get_unit_by_pid_cgroup(m, pid);
2012 #if 0 /// elogind must substitute this with its own variant
2013 int manager_notify_cgroup_empty(Manager *m, const char *cgroup) {
2019 log_debug("Got cgroup empty notification for: %s", cgroup);
2021 u = manager_get_unit_by_cgroup(m, cgroup);
2025 return unit_notify_cgroup_empty(u);
2028 int manager_notify_cgroup_empty(Manager *m, const char *cgroup) {
2034 log_debug("Got cgroup empty notification for: %s", cgroup);
2036 s = hashmap_get(m->sessions, cgroup);
2039 session_finalize(s);
2042 log_warning("Session not found: %s", cgroup);
2047 #if 0 /// UNNEEDED by elogind
2048 int unit_get_memory_current(Unit *u, uint64_t *ret) {
2049 _cleanup_free_ char *v = NULL;
2055 if (!u->cgroup_path)
2058 if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0)
2061 r = cg_all_unified();
2065 r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v);
2067 r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v);
2073 return safe_atou64(v, ret);
2076 int unit_get_tasks_current(Unit *u, uint64_t *ret) {
2077 _cleanup_free_ char *v = NULL;
2083 if (!u->cgroup_path)
2086 if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0)
2089 r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v);
2095 return safe_atou64(v, ret);
2098 static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) {
2099 _cleanup_free_ char *v = NULL;
2106 if (!u->cgroup_path)
2109 r = cg_all_unified();
2113 const char *keys[] = { "usage_usec", NULL };
2114 _cleanup_free_ char *val = NULL;
2117 if ((u->cgroup_realized_mask & CGROUP_MASK_CPU) == 0)
2120 r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", keys, &val);
2124 r = safe_atou64(val, &us);
2128 ns = us * NSEC_PER_USEC;
2130 if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0)
2133 r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v);
2139 r = safe_atou64(v, &ns);
2148 int unit_get_cpu_usage(Unit *u, nsec_t *ret) {
2154 /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was
2155 * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply
2156 * call this function with a NULL return value. */
2158 r = unit_get_cpu_usage_raw(u, &ns);
2159 if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) {
2160 /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our
2164 *ret = u->cpu_usage_last;
2170 if (ns > u->cpu_usage_base)
2171 ns -= u->cpu_usage_base;
2175 u->cpu_usage_last = ns;
2182 int unit_reset_cpu_usage(Unit *u) {
2188 u->cpu_usage_last = NSEC_INFINITY;
2190 r = unit_get_cpu_usage_raw(u, &ns);
2192 u->cpu_usage_base = 0;
2196 u->cpu_usage_base = ns;
2200 bool unit_cgroup_delegate(Unit *u) {
2205 c = unit_get_cgroup_context(u);
2212 void unit_invalidate_cgroup(Unit *u, CGroupMask m) {
2215 if (!UNIT_HAS_CGROUP_CONTEXT(u))
2221 /* always invalidate compat pairs together */
2222 if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO))
2223 m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO;
2225 if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT))
2226 m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT;
2228 if ((u->cgroup_realized_mask & m) == 0)
2231 u->cgroup_realized_mask &= ~m;
2232 unit_add_to_cgroup_queue(u);
2235 void manager_invalidate_startup_units(Manager *m) {
2241 SET_FOREACH(u, m->startup_units, i)
2242 unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO);
2245 static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = {
2246 [CGROUP_AUTO] = "auto",
2247 [CGROUP_CLOSED] = "closed",
2248 [CGROUP_STRICT] = "strict",
2251 DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy);